Open wangm-word opened 3 years ago
generate a mask applying on Z, see https://github.com/JimmySuen/integral-human-pose/blob/ad3f875bc05538da3471ef81484e23fad3e9c787/pytorch_projects/common_pytorch/dataset/mpii.py#L171
in this way, loss of z won't affect BP.
thank you for your great work! I want to know,when you train it ?the 2D data miss "Z" coordinate value,how to deal with it?